Method, apparatus, and medium for video processing

Abstract

Embodiments of the present disclosure provide a solution for video processing. A method for video processing is proposed. The method comprises: determining, during a conversion between a target video block of a video and a bitstream of the video, a target coding tool for the target video block by using a machine learning model; and performing the conversion by using the target coding tool. By taking the machine learning model into consideration in selecting the coding tool, a more proper coding tool can be selected. In this way, the coding performance can be enhanced.

I/We claim: 
     
         1 . A method for video processing, comprising:
 determining, during a conversion between a target video block of a video and a bitstream of the video, a target coding tool for the target video block by using a machine learning model; and   performing the conversion by using the target coding tool.   
     
     
         2 . The method of  claim 1 , wherein the machine learning model is used for neural network (NN) filtering during the determination of the target coding tool,
 wherein the machine learning model is obtained by an encoder used during the conversion,   wherein determining the target coding tool comprises: applying the machine learning model in a rate-distortion optimization (RDO) process on the target video block to obtain the target coding tool,   wherein the machine learning model is not obtained by a decoder used during the conversion, or   wherein the machine learning model comprises at least one of: a neural network (NN) model, a convolutional neural network (CNN) model, or a non-NN based model.   
     
     
         3 . The method of  claim 1 , wherein a further model different from the machine learning model is obtained by an encoder used during the conversion,
 wherein the machine learning model is combined with the further model,   wherein the further model comprises at least one of: a convolutional neural network (CNN) model, a deblocking filter, a sample adaptive offset (SAO) filter, an adaptive loop filter (ALF), a cross-component SAO (CCSAO) filter, or a cross-component ALF (CCALF).   
     
     
         4 . The method of  claim 3 , further comprising:
 applying, based on a filtering order, at least one of the machine learning model or the further model to the target video block,   wherein the filtering order comprises an order as follows: a deblocking filter, a convolutional neural network (CNN) filter, a sample adaptive offset (SAO) filter, and an adaptive loop filter (ALF),   wherein the filtering order is predefined.   
     
     
         5 . The method of  claim 1 , wherein determining the target coding tool comprises:
 determining filtered reconstruction information of the target video block by using the machine learning model; and   determining the target coding tool based on the filtered reconstruction information.   
     
     
         6 . The method of  claim 1 , wherein determining the target coding tool by using the machine learning model comprises at least one of:
 determining a target intra mode by using the machine learning model;   determining a target coded intra tool by using the machine learning model;   determining a target inter mode by using the machine learning model;   determining a target coded inter tool by using the machine learning model;   determining a target partitioning mode by using the machine learning model;   determining a target transform core by using the machine learning model; or   determining a target coded tool by using the machine learning model,   wherein determining the target partitioning mode comprises: determining the target partitioning mode from a quad-tree (QT) partitioning mode, a binary-tree (BT) partitioning mode, a ternary-tree (TT) partitioning mode, or a non-split mode.   
     
     
         7 . The method of  claim 1 , wherein determining the target coding tool by using the machine learning model comprises:
 determining a second distortion of the target video block based at least in part on the machine learning model; and   determining the target coding tool based on the second distortion,   wherein the second distortion comprises a cost of the target video block, or   wherein the second distortion is determined with one of: a sum of square error (SSE) matrix, a mean of square error (MSE) matrix, a structural similarity (SSIM) matrix, a multi-scale structural similarity (MS-SSIM) matrix, or an information content weighted SSIM (IW-SSIM) matrix.   
     
     
         8 . The method of  claim 7 , wherein determining the second distortion comprises:
 determining, based on reconstruction samples of the target video block, first filtered reconstruction samples of the target video block by using the machine learning model, the reconstruction samples being reconstructed without an in-loop filtering; and   determining the second distortion based on a third distortion between the first filtered reconstruction samples and original samples of the target video block.   
     
     
         9 . The method of  claim 8 , wherein determining the second distortion based on the third distortion comprises: determining the third distortion as the second distortion without using a fourth distortion between the reconstruction samples and the original samples of the target video block, or
 wherein determining the second distortion based on the third distortion comprises: determining the second distortion based on the third distortion and a fourth distortion between the reconstruction samples and the original samples of the target video block, or   wherein determining the second distortion based on the third and fourth distortions comprises: determining a first metric of the third and fourth distortions as the second distortion.   
     
     
         10 . The method of  claim 7 , wherein determining the second distortion comprises:
 determining, based on a set of reconstruction samples of the target video block, at least one set of filtered reconstruction samples of the target video block by using at least one filtering model;   determining a fifth distortion between the set of reconstruction samples and a set of original samples of the target video block;   determining at least one distortion between the at least one set of filtered reconstruction samples and the set of original samples of the target video block; and   determining the second distortion based on the fifth distortion and the at least one distortion,   wherein the at least one filtering model comprises at least one of: the machine learning model, or a further model different from the machine learning model, or   wherein determining the second distortion based on the fifth distortion and the at least one distortion comprises: determining a second metric of the fifth distortion and the at least one distortion as the second distortion.   
     
     
         11 . The method of  claim 1 , wherein a filtering process is applied to reconstruction samples of the target video block by using the machine learning model during the determination of the target coding tool,
 wherein the filtering process is different from an in-loop filtering process or a post-processing process applied to the target video block,   wherein the machine learning model used in the filtering process is different from a further filtering model used in the in-loop filtering process or the post-processing process,   wherein a first number of machine learning models used in the filtering process is different from a second number of further filtering models used in the in-loop filtering process or the post-processing process,   wherein a first network structure of the machine learning model is different from a second network structure of the further filtering model,   wherein the filtering process is applied to a sub-region of the target video block,   wherein the sub-region of the target video block comprises at least one of: boundary samples of the target video block, or inner samples of the target video block, or   wherein the filtering process is applied to a down-sampled version of the target video block.   
     
     
         12 . The method of  claim 1 , wherein the machine learning model is the same as a further machine learning model obtained by a decoder used during the conversion,
 wherein a first number of residual blocks of the machine learning model is the same as a second number of residual blocks of the further machine learning model.   
     
     
         13 . The method of  claim 1 , wherein the machine learning model is different from a further machine learning model obtained by a decoder used during the conversion,
 wherein the machine learning model is simpler than the further machine learning model, or   wherein a first depth of the machine learning model is different from a second depth of the further machine learning model, or   wherein the first depth is shallower than the second depth, or   wherein a first feature map of the machine learning model is different from a second feature map of the further machine learning model, or   wherein a first number of feature maps of the machine learning model is less than a second number of feature maps of the further machine learning model, or   wherein a first number of residual blocks of the machine learning model is different from a second number of residual blocks of the further machine learning model, or   wherein the first number of residual blocks of the machine learning model is less than the second number of residual blocks of the further machine learning model, or   wherein a first convolution kernel of the machine learning model is different from a second convolutional kernel of the further machine learning model.   
     
     
         14 . The method of  claim 1 , further comprising:
 determining second information regarding the machine learning model based on coding information of the target video block, or   wherein the second information comprises at least one of:   whether to use the machine learning model in the determination of the target coding tool, or   how to use the machine learning model in the determination of the target coding tool.   
     
     
         15 . The method of  claim 14 , wherein the coding information comprises at least one of: a coding mode of the target video block, or coding statistics of the target video block, or
 wherein the coding information comprises at least one of: a prediction mode of the target video block, a quantization parameter (QP) of the target video block, a temporal layer of the target video block, or a slice type of the target video block, or   wherein the coding information comprises at least one of: a quantization step in the conversion, a temporal layer in the conversion, a slice type of the target video block, a block size of the target video block, a color component of the target video block, or a rate distortion cost of the target video block without using the machine learning model.   
     
     
         16 . The method of  claim 1 , wherein the conversion includes encoding the target video block into the bitstream. 
     
     
         17 . The method of  claim 1 , wherein the conversion includes decoding the target video block from the bitstream. 
     
     
         18 . An apparatus for processing video data comprising a processor and a non-transitory memory with instructions thereon, wherein the instructions upon execution by the processor, cause the processor to:
 determine, during a conversion between a target video block of a video and a bitstream of the video, a target coding tool for the target video block by using a machine learning model; and   perform the conversion by using the target coding tool.   
     
     
         19 . A non-transitory computer-readable storage medium storing instructions that cause a processor to perform a method performed by a video processing apparatus, wherein the method comprises:
 determining, during a conversion between a target video block of a video and a bitstream of the video, a target coding tool for the target video block by using a machine learning model; and   performing the conversion by using the target coding tool.   
     
     
         20 . A non-transitory computer-readable recording medium storing a bitstream of a video which is generated by a method performed by a video processing apparatus, wherein the method comprises:
 determining, by using a machine learning model, a target coding tool for a target video block of the video; and   generating the bitstream by using the target coding tool.